Research map / Label-efficient and robust learning / Recognition and long-tail
OpenFWI: Large-scale Multi-structural Benchmark Datasets for Full Waveform Inversion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| VelocityGAN vs InversionNet | VelocityGAN · 2D OpenFWI datasets | SSIM: 0.9461 -> 0.9521 (FlatVel-B); 0.6727 -> 0.7111 (CurveVel-B) | Helped |
| VelocityGAN vs InversionNet training time | VelocityGAN · Vel family | training time (h): 2h -> 8.6h | Hurt |
| unsupervised UPFWI vs supervised methods | UPFWI · 2D OpenFWI datasets | SSIM: 0.9895 -> 0.9565 (FlatVel-A, vs InversionNet); 0.6163 -> 0.3941 (CurveFault-B) | Hurt |
| l2 loss instead of l1 loss (ablation) | VelocityGAN · FlatVel-A | MAE: 0.0118 -> 0.0605 | Hurt |
| l2 loss instead of l1 loss (ablation) | VelocityGAN · FlatFault-A | SSIM: 0.9313 -> 0.9798 | Helped |
| l2 loss instead of l1 loss (ablation) | InversionNet · Style-B | SSIM: 0.6314 -> 0.7599 | Helped |
| l2 loss instead of l1 loss (ablation) | InversionNet · FlatVel-A | MAE: 0.0131 -> 0.0111 | Helped |
| l2 instead of l1 loss (ablation) | InversionNet3D · 3D Kimberlina-V1 | SSIM: 3% SSIM drop | Hurt |
| higher velocity map complexity (harder datasets) | InversionNet/VelocityGAN/UPFWI · Vel and Fault families | SSIM | Hurt |
| train on 2D slices instead of 3D network (ablation) | InversionNet · 3D Kimberlina-V1 | SSIM: 0.9652 vs 0.9838 | Mixed |
| add noise to test seismic data (ablation) | InversionNet · 2D datasets | accuracy | Hurt |
Closest papers by meaning
- A Unified Framework for Forward and Inverse Problems in Subsurface Imaging using Latent Space Translations (ICLR 2025 · similarity 0.89)
- Auto-Linear Phenomenon in Subsurface Imaging (ICML 2024 · similarity 0.86)
- A Novel Diffusion Model for Pairwise Geoscience Data Generation with Unbalanced Training Dataset (AAAI 2025 · similarity 0.85)
- EasyInv: Toward Fast and Better DDIM Inversion (ICML 2025 · similarity 0.85)
- InverseBench: Benchmarking Plug-and-Play Diffusion Priors for Inverse Problems in Physical Sciences (ICLR 2025 · similarity 0.84)
- FreeInv: Free Lunch for Improving DDIM Inversion (NeurIPS 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.